TL;DR

On Tuesday 6 October the government said it will act on all 44 proposals from the National Commission on regulating healthcare AI, a doctor-led body the MHRA set up, whose report came out on 10 September. The MHRA has opened applications for AI Airlock’s third phase. The sandbox for AI medical devices will now focus on how products are watched after they reach patients. A full delivery roadmap is promised by spring 2027.

What the response commits to

The headline shift is away from approving an AI device once and towards checking it for as long as it is in service. The Commission argued that a single assessment tells regulators too little about how a tool performs in real clinical settings, and ministers have adopted that view.

Early deliverables are concrete. The MHRA says it will issue draft guidance, due by December 2026, on how manufacturers should handle updates to AI devices that change over time. Next year it will consult on how such products are qualified and classified, which was the Commission’s first recommendation. Ministers will also look at staged authorisation, letting promising tools reach NHS use sooner under close supervision while real-world evidence builds up.

A newly formed Programme Board will drive the work across nine priority areas, with progress updates after the spring roadmap.

AI Airlock moves to post-market surveillance

AI Airlock has won three more years of government funding after its pilot and second phase. Phase 3 concentrates on surveillance after launch and on regulation across a product’s working life. A webinar for applicants runs on 22 October, applications close in November, and the first cohort is chosen that month.

James Frith, the health innovation minister, said innovation “must never come at the expense of patient safety”. Professor Alastair Denniston, who chaired the Commission, said ministers had listened to “more than 12,000 patients, clinicians and members of the public” who fed into the work.

Looking forward

In our view, accepting everything is the easy part. The lifecycle model, which we examined in our analysis of the Commission’s L-plate approach, depends on evidence gathered after deployment, and who collects it is a question the spring roadmap will have to settle. For UK health-tech firms, the December change-management guidance is the first test of how demanding that will be, and Scotland has said it will publish its own response in due course. Suppliers selling AI tools into the NHS should treat the Airlock’s third phase as an early view of the evidence regulators will expect.